01
今日三件事
跨扫描重复出现的大信号
商业 · 亏损
Anthropic 上半年净亏损 $80.7B：同题三连发Anthropic disclosed an $80.7B net loss for the first half of 2026
同题连发三帖，口径精确到 $80.7B；亏损构成 raw 未载。
商业 · 竞争
DeepSeek 传融资 $15B，叫板 OpenAI 与 AnthropicDeepSeek just raised $15 billion to take on OpenAI and Anthropic
两路扫描同日命中同一标题；轮次、出资方与估值 raw 一概未载。
安全
用 Claude 收缩代码的攻击面Using Claude to reduce the attack surface of code
今日大热讨论（134 pts）；与形式化验证帖同日呼应，AI 正被用于压低代码风险面。
02
模型与 Agent 工具潮
CLI 领衔，工具小件连发A1
OpenAI CLI 冲上热榜OpenAI CLI
2026-10-22 · 229 pts / 109 评纯标题条目，是否官方发布 raw 无载。
A2
热问：模型这么强，OpenAI 为何不直接开 100 家公司Ask HN: If the models are so good, why doesn't OpenAI just start 100 companies?
2026-10-22 · 55 pts / 60 评两路扫描同日命中，60 评讨论模型厂商的边界。
A3
CATT 0.2.1：在 CI 里测 agent、对标 Muse/Instinct 类Show HN: CATT 0.2.1: Test agents/services against a Muse/Instinct-alike, in CI
2026-10-22 · 5 pts / 0 评两路扫描同日命中的 CI 测试工具；对标细节 raw 无载。
A4
Halv：开源 agent 编码工具，主打省用量Show HN: Halv – open-source agent coding software that reduces your usage
2026-10-22 · 6 pts / 2 评开源、主打省用量；实测数据 raw 无载。
03
钱与市场
预测市场快照，结构市场优先核验说明：本表只收问题与选项可一一对应、且概率/趋势/成交量三件套齐备并较上期可见变化的市场行；Anthropic 破产盘（2027 档 10% 与上期持平）与 OpenAI 收盘市值盘（43% 持平）本期不列，Coding Arena（$16K）、LiveBench（$5K）与 Anthropic 收盘市值中段盘（$79K）量级过小，一并从略。
| 预测市场问题 | 概率 | 趋势 | 事件成交量 |
|---|---|---|---|
| 【进展】Anthropic 12 月底前 IPO？升至 87%（上期 86%）Will Anthropic IPO by December 31, 2026? | 87% | 周 +12.5% | $1.4M |
| 【进展】Anthropic IPO 落在 11 月？档位升至 73%（上期 71%）Will Anthropic's initial public offering occur in November 2026 (ET)? | 73% | 月 +17.7% | $153K |
| 【进展】Anthropic IPO 收盘市值 $2.0T–$2.25T 档回落至 18%，$1.75T–$2.0T 档 28% 领先Will Anthropic's market cap be between $2.0T and $2.25T at market close on IPO day? | 18% | 月 −5.9% | $1.2M |
| 【进展】OpenAI 年内上市？12 月底档仅 4%，最早档押 2027-06（34%）Will OpenAI IPO by December 31 2026? | 4% | 今日 −1.1% | $125K |
| 【进展】OpenAI IPO 估值低于 $1T？档位 26%，与下调估值报道同向Will OpenAI's IPO valuation be less than $1T? | 26% | 月 +8.0% | $230K |
| 【进展】10 月底最强模型归 Google？升至 89%（上期 81%）Will Google have the best AI model at the end of October 2026? | 89% | 月 +81.0% | $2.6M |
| 【进展】年底最强模型盘：Anthropic 降至 46%，Google 档 45% 贴身Will Anthropic have the best AI model at the end of December 2026? | 46% | 月 −27.0% | $1.3M |
| 【进展】最强中国模型盘：小米档 69% 领跑，阿里档月内 −55 个百分点Will Alibaba have the best Chinese AI model at the end of October 2026? | 10% | 月 −55.0% | $137K |
| 【进展】OpenAI 再解千禧年大奖？2027 档跌至 52%（上期 50%）Will OpenAI announce another Millennium Prize solution by December 31, 2027? | 52% | 月 −30.0% | $604K |
| 【进展】年内 AI 零新增千禧年难题？升至 81%（上期 78%）Will AI solve 0 more Millennium Prize Problems in 2026? | 81% | 月 +31.5% | $254K |
M1
OpenAI 据报下调 IPO 估值目标OpenAI trims valuation target after pitching higher price
2026-10-22 · 3 pts / 1 评与「估值低于 $1T」档 26%（月 +8.0%）、年内上市档 4% 同向。
04
安全 · 治理 · 版权
形式化验证 × 治理动作R1
LLM 让形式化软件验证走向大规模LLMs are making formal software verification feasible at a much larger scale
2026-10-22 · 41 pts / 11 评与头条「收缩攻击面」帖同日成对信号；工具与案例 raw 无载。
R2
内部见闻：Anthropic 推向递归自我改进What I saw inside Anthropic: AI lab pushes towards recursive self-improvement
2026-10-21 · 7 pts / 2 评仅标题级信号，正文佐证 raw 无载。
R3
Anthropic 将 Agent Skills 捐给 Agentic AI 基金会Donating Agent Skills to the Agentic AI Foundation
2026-10-22 · 5 pts / 0 评生态治理动作，条款与受让细节 raw 未载。